ABSTRACT:

We study the following drawback: A data distributor has given sensitive information to a group of supposedly trusted agents (third parties). Some of the information are leaked and found in an unauthorized place (e.g., on the internet or someone's laptop). The distributor must assess the probability that the leaked knowledge came from one or more agents, vs having been independently gathered by different means that. We have a tendency to propose data allocation ways (across the agents) that improve the chance of identifying leakages. These methods don't rely on alterations of the released information (e.g., watermarks). In some cases, we have a tendency to can additionally inject “realistic but faux” knowledge records to further improve our probabilities of detecting leakage and identifying the problem.